I design and ship machine-learning systems end to end — from Transformer surrogate models on quantum-network data to production RAG pipelines and IoT at the edge.
I'm a Master's student in Computer Science on the Artificial Intelligence and Machine Learning track at Binghamton University's Thomas J. Watson College. My work sits at the intersection of production ML engineering and applied research — shipping systems that hold up in the real world, then measuring them honestly.
Right now I'm integrating LLMs into a multi-tenant voice platform at Zoo Media and keeping a campus AI system running for ~24,000 students and staff. I recently completed research on Transformer-based surrogate models for quantum networks. Across all of it, I care about the same thing: real systems, real metrics, and being able to defend every number.
Earlier, I completed my BE in Computer Science & Engineering at Savitribai Phule Pune University - AISSMS College of Engineering, Pune, and published four papers across applied ML, NLP, and IoT.
A selection across production ML, GenAI, and embedded systems. Every metric below is measured, not estimated.
Production agentic RAG that ingests arXiv papers daily and answers grounded research questions. SHA-256 Redis caching over a hybrid BM25 + vector RRF pipeline; deployed to AWS EC2 at $0.
13-week cashflow forecasting with grounded natural-language Q&A and accounting-invariant enforcement — built as a production-style service, not a toy notebook.
In-browser bidirectional sign-language translation across 40+ languages, 543 landmarks/frame at sub-second WebGPU latency. A re-implementation inspired by the open-source sign.mt project.
Reproduces and extends Ishida et al. (ICLR 2023), stress-testing soft-label Bayes-error estimation under annotation bias and miscalibration, with temperature-scaling calibration on CNN variants.
Stacked ensemble (XGBoost + LightGBM + CatBoost) with SHAP interpretability and a FastAPI deployment — modeling plus explainability and serving.
Sub-$100 private NAS on a Raspberry Pi 4 with cross-platform SMB/CIFS access. Benchmarked four storage tiers and diagnosed Gigabit Ethernet — not disk speed — as the true network bottleneck.
NFC authentication plus LoRaWAN tracking with a lock-state escalation model and a Flask dashboard, validated end-to-end over The Things Network.
Automated voice-agent testing harness — 10 patient personas, Groq Whisper transcription, heuristic transcript analysis, and structured bug reports replacing manual call testing.
Four peer-reviewed papers in IJRASET across applied ML, NLP, and IoT systems.
I'm looking for ML / AI Engineer roles. If you're hiring or just want to compare notes on RAG systems or quantum ML, reach out.